The interest in AI has skyrocketed thanks in large to the rise of new AI models that have the potential to change the nature of industries on a global scale. With how data intensive AI is, its soaring popularity has also kickstarted a potential billion-dollar spending campaign on data centers. Let’s explore the basics of AI, its rising popularity, and how the data center drives the next generation of AI.
Background on AI
A common question/talking point for the general population when discussing artificial intelligence, is simply: “What is Artificial Intelligence?” To many people, AI should behave like a human-like robot, akin to something out of an 80's or 90's science fiction movie. Others, particularly AI developers, tend to think of AI as an advanced tool to solve problems. We let an AI model answer this question itself:
Question:
What is Artificial Intelligence?
Answer from AI:
Artificial Intelligence (AI) is the ability of machines to perform tasks that would normally require human intelligence such as visual perception, speech recognition, decision-making, and language translation. AI is a set of technologies that enable computers to perform a variety of advanced functions, including the ability to see, understand and translate spoken and written language, analyze data, make recommendations, and more. AI is an academic discipline with various views on what constitutes its area of research, goals and approaches used.
Artificial Intelligence is a very broad term that encompasses a wide range of categories and applications. Because the expectations for AI range from the practical to the more ambitious, AI models are often categorized based on their applications.
Generative AI
Recent advancements in AI have been in a subcategory of AI called Generative AI. This is a type of AI that is able to produce content that ranges from text, images, video, audio, and 3D models, based on prompts. In other words, the AI model can take an input of questions or instructions and output content with that information.
Generative AI models are able to accomplish this through the utilization of transformer models, which are artificial neural networks that can “learn” context. Transformer architectures contain both an encoder and a decoder. The encoder takes a question or set of instructions and simplifies the question/instructions to make them more concise. The decoder then takes this simplified version and uses an algorithm called Self-Attention to assign weighted values to each word based on its importance to the sentence.
Take the sentence “What are the best restaurants around this area?” for example. The AI model can weigh the terms ‘restaurants’ ‘best’ and ‘around’ more heavily compared to the other words. The model then highlights ‘restaurants’ as the main topic of the sentence and works from there, enabling it to “understand” the context of the question. The model is able to assign relatively accurate weighted values to words because it has been pre-trained on a massive amount of data from the internet, including trillions of text, code, etc. It then uses word prediction based on the previous context to generate additional text.
Evolving AI
So, what is the hype about current AI models all about? Why have recent AI models generated so much interest?
Much of the hype around these AI models stems from their ability to tackle complex tasks such as writing emails, essays, coding/debugging and more. Generative AI models are able to produce content with high coherency, which would normally require a human to do so. To illustrate their potency, here we demonstrate what current iterations of AI are able to do with simple prompts.
As a follow-up instruction to the original question that was asked earlier, we asked the AI model to explain to a kid what AI is.
Artificial Intelligence (AI) is when computers and machines can do things that normally require human intelligence like learning, seeing, hearing, and decision-making. It’s like having a robot friend who can help you with your homework or play games with you. Just like how you learn new things every day, AI can also learn new things every day by looking at pictures and listening to sounds. AI is really smart and can help people do things faster and better than before.
In addition to producing conversational-like text with high coherency and being able to adjust the content for different audiences, AI models can also generate photos or photoshop to an impressive degree. Compare and contrast original photos and AI photos. For example:
Same person outside in a field
In this example, the AI model is able to take the photo of a person in a data center and transpose them over a background of a baseball field while adding a baseball glove to his hands.
Here is another example:
Generative fill the other parts of the photo
In this example, the AI model can generative fill a photo by "filling out" the rest of the photo despite having no external details of the scene of the photo. It predicts what it would look like if the camera zoomed outwards.
What we are seeing here are very early versions of what AI is capable of, but as one can see, even in their early forms, AI models are able to generate high quality content. With more refining, these AI models should continue to evolve.
Concerns regarding AI
Despite all the hype surrounding AI, it also comes with its fair share of controversy. The topic itself is polarizing and has brought about major concerns. It has been widely praised and scrutinized due to its potential to impact and even disrupt a wide range of industries.
Despite all the hype surrounding AI, it also comes with its fair share of controversy. The topic itself is polarizing and has brought about major concerns. It has been widely praised and scrutinized due to its potential to impact and even disrupt a wide range of industries.
Generative AI models have also generated controversy with ethical concerns. These concerns range from cheating in academia, to deep faking content for malicious purposes, to generating discriminatory content stemming from implicit biases. When taking these ethical issues into consideration, there are legitimate questions on how much regulation is needed. Issues like the aforementioned AI hallucination flaw can also impact an organization's ability to produce accurate and appropriate content/messaging without proper moderation.
Relationship between AI and the data center
While the discussion surrounding AI is often contentious, one thing that is widely agreed upon is the importance of the data center for AI. The data center serves as the physical engine that drives AI and will need to be heavily optimized to run next-generation AI applications. Data center components that will be a key focus include:
Memory and storage
Machine learning is very compute intensive as it utilizes algorithms that require ultra-efficient processing. This along with the fact that machine learning models leverage large sets of data to improve performance, means that reliable memory and storage components will be high in demand. Pairing high-bandwidth, low-latency memory with faster performance, high IOPS and large capacity storage devices will be essential in powering the data center to handle some of the most rigorous AI workloads. AI models can also attest to this. In the words of AI models:
Memory and storage are critical elements to AI development. AI accelerators used to train AI models in data centers require the highest memory bandwidth available. AI applications have high storage capacity demands that can easily start in the terabyte range and scale into hundreds of petabytes. Smarter AI need compute, memory (lots of storage and memory) and connectivity.
Networking equipment
One of the main bottlenecks for AI is network performance. Without gigabit plus connectivity, the potential of AI performance is capped. Thus, advanced optical transceivers that can efficiently transport data back and forth between edge data centers will be of utmost importance. This notion is backed by AI:
Optical transceivers are crucial in developing better AI systems by facilitating the rapid, reliable data transmission these systems need to do their jobs. High-speed, high-bandwidth connections are essential to interconnect data centers and supercomputers that host AI systems and allow them to analyze a massive volume of data. To work optimally, they need information from all network layers and ever-faster data centers to process it quickly. Pluggable optical transceivers thus need to become smarter, relaying more information back to the AI central unit, and faster, enabling increased AI processing.
Mutually beneficial relationship
As much as the future of AI depends greatly on the prowess of the data center, the relationship between AI and the data center is also symbiotic. AI can greatly bolster data centers and optimize how data centers are run. As AI puts it:
AI can help optimize data center performance by improving capacity planning and resource allocation, improving data center security, getting in front of carbon footprint reduction and reporting, improving uptime maintenance levels and benchmarking data center performance over time. AI can also automate cooling, electricity, power and security systems. Incorporating AI into the data center is becoming increasingly necessary for every data-driven business.
Businesses have already started to ramp up the investment in data centers. Industry experts project that generative AI data center infrastructure and its operating costs will exceed the $76 billion threshold by the year 2028. The massive increase in spending is a testament to the growing potential of AI.
Axiom powers the data center
Axiom has been a trusted, third-party provider of high-value and performance data center equipment for over 25 years. With OEM equivalent performance at superior value, our full lineup of data center solutions from memory components to storage devices to optical transceivers and more, enables businesses to build performance-oriented data centers with greater accommodation for hyper-scaling. Optimize data center performance for next generation AI today with Axiom data center solutions.
Sources
https://platform.openai.com/docs/models/overview
https://www.computerweekly.com/feature/Storage-requirements-for-AI-ML-and-analytics-in-2022
https://www.computer.org/csdl/magazine/co/2022/10/09903869/1H0G6xvtREk
https://time.com/6271657/a-to-z-of-artificial-intelligence/
https://openai.com/research/language-unsupervised
https://help.openai.com/en/articles/6825453-chatgpt-release-notes
https://openai.com/blog/chatgpt
BingChat. (2023). BingChat (July 15 version) [Large language model]. Bing Chat | Microsoft Edge
Adobe Photoshop. (2023). Adobe Photoshop (July 30 version) [Generative AI]. Adobe Photoshop | Adobe
